agents and services that serve your customers
push your agent or your api, and spredr keeps it answering.
your agents, answering every customer at once
an ai agent is a service like any other, asleep at 0 cpu until someone writes. each customer gets their own queue and memory, and a risky action waits for your ok.
one task, end to end
your code does the work. spredr brings the queue, the memory, the questions and the trace.
a task arrives
from your backend, a schedule, a webhook or you. with its key, it never enters twice.
it waits its turn
every customer has their own queue: one with a thousand tasks never holds back the rest.
your agent does it
with that customer's memory and what it already learned. the model is yours, with your key.
it asks first
what is risky waits for your ok. then it goes on where it was, doing nothing twice.
the trace stays
every step with its tokens and its cost. your corrections become lessons.
your usual services, in the same place
apis, sites, workers and bots: push your repo, spredr builds it, gives it an address and keeps it answering, even when a machine falls. the agent and its backend, side by side. and when a service keeps files, its persistent disk is encrypted and backed up every 15 minutes.
a deploy your customers never notice
serving today
your customers reach v14 through spredr. every request goes to it, and its answer comes back.
you push
a commit to the branch you chose. v14 keeps serving as if nothing happened.
built and signed
v15 is built on a machine made only for builds, signed, and started beside v14. nobody reaches it yet.
checked
v15 has to answer on /health before a single customer does.
switched
spredr sends everyone to v15 at once. v14 finishes what it was serving, then stops, kept for a one-tap rollback.
it crashes
v15 starts crashing in its first minutes. spredr goes back to v14 on its own.
a machine falls
the machine falls. nothing answers there anymore, and spredr notices.
rescued
in about four minutes, spredr starts your service on another machine, checks it on /health, and sends your customers there.
back home
when the first machine is back, your service returns home: the new copy starts before the old one stops, without a cut.
from a repo to an address, in four steps
each step with a scene of its own, showing what spredr really does today.
your code gets an address
your code, an api, a site or an ai agent, lives in a repo on github. spredr takes it and gives it an address on the internet, like my-api.onspredr.com, with https already on. your customers’ phones and browsers talk to that address.
you connect it once
install the spredr app on github, pick the repo, and tell spredr two things: the port your app listens on, and a path that answers when it is ready. that is all the setup there is.
spredr builds it for you
no dockerfile needed: spredr recognizes your language and builds your code on a machine made only for builds. what comes out carries a signature, and your machines check it before they start it.
every push is a new version
each push to your branch makes a new version, and the old ones are kept. if the new one is wrong, you go back in seconds, without building again.
works hot. rests without switching off.
measured on gvisor: a service standing still went from 222 MB to 6 MB without switching off. if you choose it, it can also stop with no visits.
a machine falls. your service keeps going.
requests per minute, on an example night
only what you published runs
every version carries a unique signature, and your machines check it before starting anything.
signed, or it stays out.
not even someone inside our servers could slip in their code: your machines refuse it, and it is logged.
each one in its own box.
what happens in one service never touches the rest, and nobody gets into your machines, not even us.
what we don't do, yet
so you know before you bring a service over, and what you do instead today.
questions you would ask
yes, during the private beta. you will have it answering in minutes.
apis, sites, workers and ai agents, from your repo or your terminal. all at once, on one plan.
we watch every machine all the time. if one falls, your service answers again from another one in about four minutes, without you doing a thing.
in the united states or in europe. we recommend where it should live, and if you change your mind, it moves without going down.
push it today. sleep through the night.
spredr sends your customers only to versions that answer, and to another machine if one falls.